DNA Repair Gene XPD Polymorphisms and Cancer Risk: A Meta-analysis Based on 56 Case-Control Studies

Many environmental factors, such as radiation, diet, smoking, and endogenous or exogenous estrogens, are associated with DNA damage. Unrepaired or misrepaired DNA results in gene mutations, chromosomal alterations, and genomic instability. Several studies have suggested that genes involved in DNA repair system play a crucial role in protecting against mutations. Patients with certain cancers have reduced DNA repair proficiencies (1). Similarly, the enzymes of the nucleotide excision repair (NER) pathway have been implicated in cancer (2).

The excision repair cross-complementing group 2 (ERCC2) gene (formerly named XPD) maps to chromosome 19q13.3 and is composed of 23 exons (3). Its protein is 761 amino acids in length. The DNA repair process and gene transcription are coupled via activity of the TFIIH complex, a multiprotein complex with functions including transcription, NER, transcription-coupled repair, apoptosis, and cell cycle regulation. The XPD protein is involved in transcription-coupled NER and is an integral member of the basal transcription factor BTF2/TFIIH complex. The XPD gene product has an ATP-dependent DNA helicase activity and belongs to the RAD3/XPD subfamily of helicases (4). Mutations in the XPD gene can result in three different disorders: xeroderma pigmentosum, trichothiodystrophy, and Cockayne syndrome (5).

The XPD gene is abundant with polymorphisms (6). The single nucleotide polymorphisms (SNP) in XPD in the HapMap and dbSNP database are adequate (7, 8). However, in coding regions, most exhibit a low frequency of heterozygosity. Only four SNPs [rs13181 (9), rs1799793 (9), rs238406 (10), and rs1052555 (11)] that have been subjects of association studies have comparably high heterozygosity frequencies. Two SNPs (rs238406 and rs1052555) have not been well studied and are not matured enough to execute a meta-analysis (10, 11). The other two SNPs, which were considered as “tagging SNP,” have been mostly described to date are ERCC2_18880_A>C (rs13181) and ERCC2_6540_G>A (rs1799793; ref. 12). rs13181 is predicted to result in a lysine-to-glutamine transition at position 751. rs1799793 would confer an aspartic acid-to-asparagine change at position 312. They are in strong linkage disequilibria with each other (13). These polymorphisms may affect different protein interactions, diminish the activity of TFIIH complexes, and alter the genetic susceptibility for cancer.

Recently, several studies investigated the role of these two polymorphisms on the risk of various cancers including breast, bladder, lung, prostate, head and neck, basal cell carcinoma, and melanoma. However, the results of these studies remain controversial. It is possible that discrepancies in findings result from variations in subjects' ethnicities or differential effects of XPD polymorphisms in different tumor. As well, most studies simply examined the effects of SNPs; few considered the interaction of environmental factors. Small sample size and missing data for genotype may have contributed to the false-positive or false-negative findings. Furthermore, hospital- and population-based case-control studies have shown inconsistent results.

With an aim of addressing inconsistencies in the findings of these studies, we did a meta-analysis based on published case-control studies to better assess the association between these two XPD polymorphisms and cancer susceptibility and to further explore the interaction of smoking.

We searched MEDLINE (U.S. National Library of Medicine) for all genetic association studies on the rs13181 and rs1799793 polymorphisms of XPD and the susceptibility of cancer published from December 2000 to January 2007 using the PubMed search engine. The search used the keywords and subject terms “XPD” or “ERCC2,” “polymorphism,” and “cancer and/or carcinoma” and was limited to English-language papers. The references of all computer-identified publications were searched. In addition, the PubMed option “Related Articles” and the publications on same topic in the reference lists of the reviewed articles were retrieved to search for potentially relevant publications.

Any human association study, regardless of sample size, was included if it met the following criteria; otherwise, it was excluded: (a) the study reported genotypic frequencies of XPD polymorphisms in unrelated cancer patients and unrelated individual controls; (b) the study investigated the association between at least one of the two polymorphisms of XPD (Asp³¹²Asn and Lys⁷⁵¹Gln) and the risk of cancer; and (c) the genotype distribution of control population must be in Hardy-Weinberg equilibrium. For the articles with same population resource or overlapping data sets, the publication reporting the largest or most recent data set was included.

Data were independently extracted by two reviewers (F.W. and D.C.) using a standardized data extraction form. In every case, discrepancies were adjudicated by a third reviewer (F.L.H.) until consensus was achieved on every item. Information was collected from each article, including author, year of publication, country of origin, type of design, selection and characteristics of cases and controls, demography, ethnicity, genotype frequency, matching condition, odds ratio (OR), and adjustment of confounders. For subjects of different ethnicities, data were extracted separately and categorized as Asian, European, and American [African American, American (White), and American (mixed)]. If racial descent was not clearly indicated, we assumed the predominant ethnicity for each country.

The two reviewers (F.W. and D.C.) independently assessed the quality of the studies selected using the quality assessment score developed for genetic association studies (14), which is based on both traditional epidemiological considerations and genetic issues (15). Total scores ranged from 0 (worst) to 12 (best). Any differences were adjudicated by the reviewer (F.L.H.).

ORs were employed to evaluate the associations between XPD polymorphisms and cancer risk. For Lys⁷⁵¹Gln, we compared the risk of cancer in the Lys/Gln heterozygote and in the variant homozygote Gln/Gln with the wild-type Lys/Lys homozygote. ORs and 95% confidence intervals (95% CI) were calculated. The statistical significance of the pool OR was determined with Z test. For the Asp³¹²Asn genotype, we evaluated the same three effects. In addition, genetic models assuming both dominant and recessive effects were used. I² was applied to assess heterogeneity between studies (16). When heterogeneity was not an issue, fixed effect model with Mantel-Haenszel method was used. Otherwise, a random effect model with Inverse variance + t method was used. In addition, subjects were categorized into different subgroups by tumor site, racial descent, and study design (if a particular tumor type includes only one published study, it was categorized in the “other cancer” group). The heterogeneities among subgroups were further analyzed.

Publication bias was investigated with funnel plot, which is used as the main graphical method. To supplement the funnel plot approach, rank correlation method suggested by Begg et al. (17) and linear regression approach proposed by Egger et al. (18) were adopted. Fisher's exact probability test determined Hardy-Weinberg equilibrium for genotype frequencies.

Analyses were done with software of SPSS (version 13.0). ORs and 95% CIs were generated by Meta-analysis with Interactive Explanations version 1.51 (19).

A total of 233 studies were identified from the PubMed database as primary candidates for our analysis; 53 of them met the inclusion criteria. In addition, we manually searched for the other three potential qualified references in PubMed, Highwire Press, ProQuest, and other databases, giving a total of 56 case-control studies for the meta-analysis.

We established a database of information extracted from each article. Table 1 indicates for each study the tumor site, first author, year of publication, country, racial descent of subjects, category of study design, genotype frequency for cases and controls, and the variant allele frequency in controls. Overall, from 56 studies with a total of 61 comparisons included 25,932 cases and 27,733 controls for the Lys⁷⁵¹Gln polymorphism, 35 comparisons included 16,781 cases and 18,879 controls for the Asp³¹²Asn polymorphism were reviewed. In 31 (55%) of these studies, newly diagnosed cancer cases were confirmed by histology or pathology, except for acute lymphoblastic leukemia (ALL), which was confirmed by diagnosed with bone marrow morphology and immunophenotype, and hepatocellular carcinoma, which was confirmed by liver biopsy or the combination of increased α-fetoprotein (>400 ng/mL). Controls in 47 (84%) studies were matched by age, gender, ethnicity, or other variables.

XPD Lys⁷⁵¹Gln. The variant Gln⁷⁵¹ allele frequencies of different ethnicities are shown in Table 2 and (Figure 1 ) (Ñ values of tests with significant difference among ethnicities are indicated). Table 3 indicates the associations between polymorphism at codon 751 and cancer risks (ORs). Stratified by tumor site, individuals of the Lys/Gln and Gln/Gln genotypes have a small esophageal cancer risk compared with the Lys/Lys genotype (for Lys/Gln versus Lys/Lys: OR, 1.34; 95% CI, 1.10-1.64; Ñ = 0.003 for Z test; I² = 0.00% for heterogeneity; for Gln/Gln versus Lys/Lys: OR, 1.61; 95% CI, 1.16-2.25; Ñ = 0.005 for Z test; I² = 27.53% for heterogeneity). Relative to the Lys/Lys genotype, Gln/Gln genotype associated with increasing the risk of lung cancer (OR, 1.21; 95% CI, 1.02-1.43; Ñ = 0.026 for Z test; I² = 10.82% for heterogeneity) and ALL (OR, 1.83; 95% CI, 1.21-2.75; Ñ = 0.004 for Z test; I² = 0.00% for heterogeneity). In different ethnicities, the Gln/Gln genotype associated with a subtle cancer risk compared with Lys/Lys genotype only in the European population (OR, 1.12; 95% CI, 1.02-1.23; Ñ = 0.017 for Z test; I² = 0.00% for heterogeneity). Based on different study designs, cancer risks (ORs) were significantly associated with Gln/Gln genotype compared with Lys/Lys genotype in both population-based (OR, 1.08; 95% CI, 1.00-1.18; Ñ = 0.052 for Z test; I² = 18.07% for heterogeneity) and hospital-based (OR, 1.21; 95% CI, 1.08-1.35; Ñ = 0.001 for Z test; I² = 8.12% for heterogeneity) case-control studies. Overall, individuals of the Gln/Gln genotype had a small cancer risk compared with the Lys/Lys genotype for all the reviewed cancers in this article (OR, 1.10; 95% CI, 1.03-1.16; Ñ = 0.003 for Z test; I² = 16.04% for heterogeneity).

XPD Asp³¹²Asn.Table 4 indicates the associations between polymorphism at codon 312 and cancer risks (ORs). Significantly increased risk for bladder cancer associated with the Asp/Asn genotype compared with wild-type homozygote Asp/Asp genotype (OR, 1.24; 95% CI, 1.06-1.46; Ñ = 0.007 for Z test; I² = 0.00% for heterogeneity). Risk for lung cancer associated with the Asn/Asn genotype compared with the Asp/Asp genotype (OR, 1.27; 95% CI, 1.04-1.56; Ñ = 0.018 for Z test; I² = 48.38% for heterogeneity). In different ethnicities, a subtle cancer risk in the American (mixed) population was revealed when comparing the Asp/Asn with Asp/Asp genotype but not in other ethnicities (OR, 1.08; 95% CI, 1.00-1.17; Ñ = 0.036 for Z test; I² = 40.13% for heterogeneity). For different designs, cancer risk was significantly associated with the Asn/Asn genotype compared with Asp/Asp genotype (OR, 1.14; 95% CI, 1.00-1.30; Ñ = 0.052 for Z test; I² = 0.00% for heterogeneity) in hospital-based case-control studies.

In this meta-analysis, 28 (50%) studies analyzed gene-gene or gene-environment interactions. Eleven of them investigated the interaction of smoking and these two polymorphisms. Because in some of these studies, the data for genotypes were given as wild-type homozygote, heterozygote plus variant homozygote, rather than as three separate genotypes, we can only calculate the ORs of the two polymorphisms in dominant model according to smoking status (smoking status was defined having ever smoked or not). As indicated by Table 5, no positive associations were observed in smokers or in nonsmokers.

There was moderate heterogeneity among the studies that described the XPD Asp³¹²Asn polymorphism (Asp/Asn versus Asp/Asp, I² = 46.53%; Asn/Asn versus Asp/Asp, I² = 42.58%; Asp/Asn + Asn/Asn versus Asp/Asp, I² = 49.80%) but not among the Lys⁷⁵¹Gln polymorphism (Lys/Gln versus Lys/Lys, I² = 2.69%; Gln/Gln versus Lys/Lys, I² = 16.04%; Lys/Gln + Gln/Gln versus Lys/Lys, I² = 20.21). Details are shown in Table 3 and 4.

The graphical funnel plots of codon 751 (Lys/Gln versus Lys/Lys, Gln/Gln versus Lys/Lys) and codon 312 (Asp/Asn versus Asp/Asp, Asn/Asn versus Asp/Asp) appear asymmetric (Figure 2). No publication bias for either codon was detected with rank correlation method and linear regression approach. The details are shown in Table 6.

Common polymorphisms in DNA repair genes may alter protein function and an individual's capacity to repair damaged DNA. Deficits in repair capacity may lead to genetic instability and carcinogenesis (1, 2). The XPD gene encodes an ATP-dependent DNA helicase involved in NER and in basal transcription as part of the transcription factor TFIIH (4). Correlation of its polymorphisms and cancer risk has been studied, but the results remain controversial. A small proportion of the published studies supported the conclusion that risk for various types of cancer such as breast, bladder, lung, melanoma, and head and neck cancer are associated with polymorphisms in XPD. However, several studies draw negative conclusions and some considered as inconclusive. Therefore, we did a systematic review and meta-analysis to clarify the relationship between polymorphisms in XPD and susceptibility to cancer.

A total of 56 case-control studies were included in this meta-analysis. In the case of the XPD Lys⁷⁵¹Gln polymorphism, after stratifying by tumor site, individuals carrying both Lys/Gln and Gln/Gln genotypes have a small risk of esophageal cancer compared with the Lys/Lys genotype. The Gln/Gln genotype was associated with increasing risk of lung cancer and ALL. Overall, individuals of the Gln/Gln genotype have a small risk for cancers included in this study compared with the Lys/Lys genotype. In the case of XPD Asp³¹²Asn compared with wild-type homozygote Asp/Asp genotype, Asp/Asn genotype associated with risk for bladder cancer. The Asn/Asn genotype associated with lung cancer. In this meta-analysis, no significant association was found for all kinds of reviewed cancers for Asp³¹²Asn polymorphism. For some subgroups, there were insufficient data to assess the effect of the Asp³¹²Asn polymorphism on cancer risk. Meanwhile, significantly associations were observed in both dominant and recessive models for Lys⁷⁵¹Gln. No association was found for either dominant or recessive models of susceptibility for Asp³¹²Asn.

Genotype frequency of various polymorphic loci may manifest racial differences. The world population falls into three distinct, basic racial groups: Mongoloid (such as Asian), Negroid (African), and Europeoid (Caucasian or White; ref. 20). Because race in the U.S. population is highly heterogeneous, it was categorized as African American, American (White), and American (mixed) in this analysis. No significant difference among allele frequencies was observed among European, American (White), and American (mixed) for both Gln⁷⁵¹ and Asn³¹². The Gln/Gln genotype was associated with cancer risk compared with the Lys/Lys genotype only in the European population. The Asp/Asn genotype was associated with cancer risk compared with the Asp/Asp genotype only in the American (mixed) population.

Due to the large number of statistical tests undertaken for this analysis, Z test Ñ values were adjusted to reduce the type I error induced by multiple tests. Because each test is considered independently from one another, Ñ value was adjusted with corresponding original α of 5% level by dividing with the number of subgroups [(Ñ_adjusted = α / m, m = 10 for the number of tumor sites and m = 5 for the number of ethnicities); data not shown]. This negated some original positive associations such as for lung cancer (Gln/Gln versus Lys/Lys, Asn/Asn versus Asp/Asp) and in the subgroups of European (Gln/Gln versus Lys/Lys) and American (mixed; Asp/Asn versus Asp/Asp). Other positive results are statistically conclusive. Overall, our meta-analysis results suggest that XPD polymorphisms are associated with cancer susceptibility to a lesser extent.

The pool ORs were calculated after stratifying by study design. As shown in Tables 3 and 4, the Gln/Gln genotype increased cancer risk significantly in both population-based and hospital-based studies but not in nested case-control studies. The Asn/Asn genotype was significantly associated with cancer risk compared with Asp/Asp genotype in hospital-based case-control studies. Generally, the genotype cannot be modified by environmental factors and the distribution of genotypes in common population is consistent with Hardy-Weinberg equilibrium. Therefore, there is no difference of genotype frequencies between the populations of prospective or retrospective studies. The discrepancies in results of different study designs may be caused by small sample sizes.

Tobacco smoke contains hundreds of chemicals, such as polycyclic aromatic hydrocarbons, aromatic amines, and N-nitroso compounds, which act as carcinogens in laboratory animals (21). Studies have shown that cigarette smoke increases the production of reactive oxygen species and induces DNA adducts (22). Tobacco smoke has been established as a risk factor for lung, head and neck, and pancreatic cancers (23). Results concerning smoking in breast cancer are controversial (24, 25). NER may be an important pathway modulating susceptibility to cancer, because it is the primary mechanism for the repair of bulky and helical distorting DNA adducts generated by tobacco smoke (26). Inconsistencies in the findings of different investigations might be due to the variability in the distribution of genotypes for NER genes in different ethnicities. In this study, no positive association is observed either in smokers or in nonsmokers. This may result from a differential effect of smoking in different cancers, as well as the weight of each study, which dictated by sample size in our meta-analysis. A true positive result maybe masked by creating pool ORs. Moreover, not all of the reviewed studies analyzed the same environmental factors such as alcohol consumption, menopausal status, occupational exposure, or family history and particularly the medical history of first-degree relatives. Therefore, except for smoking, we cannot evaluate the interactions of other environmental factors with XPD polymorphisms. Larger and rigorous analytical studies will be required to evaluate other environmental interactions.

Discrepancies in the conclusions of some studies may be explained by one or more of the following factors: First, diagnoses in several studies were not confirmed by pathology or histologic methods. Second, PCR-restriction fragment length polymorphism was used for genotyping in 39 (70%) studies and the automated TaqMan allelic discrimination assay was used in other studies, which may influence the genotyping results. Third, it is plausible that carcinogenic mechanism may differ by different tumor sites and that XPD polymorphisms may exert varying effects in different cancer. Finally, cancer is a multifactorial disease that results from complex interactions between environmental and genetic factors, which vary in different populations.

A systematic review of the association of XPD polymorphisms with cancer risk is statistically more powerful than any single study. It has been recommended that a significant polymorphism study with well-designed and reliable result must be large, favorably with at least >500 cases and controls (27). The sample sizes of some reviewed studies were too small to explore the subtle associations between polymorphisms and cancer risk, but the pool ORs generated from 56 case-control studies significantly increase the statistical power. Moreover, genotype frequencies of different ethnic populations were calculated and compared based on the cumulative data in this meta-analysis, and the ethnicity carrying the highest or lowest frequency of susceptible genotype can be assessed. Finally, no obvious publication bias was detected in this analysis. These considerations support the reliability of this meta-analysis.

However, the limitations of this meta-analysis should be kept in mind. First, the interactions among gene-gene, gene-environment, and even different polymorphism loci of the same gene may regulate gene expression, affect the function of the gene product, and lead to differing OR values in different cancers. Because lacking of the original data of reviewed studies limited our further evaluation of potential interactions. Second, SNPs that are in linkage disequilibrium constitute haplotypes with sufficient information. SNPs as markers at the linked loci can be predicted based on the known markers nearby. Additionally, the power of statistical analysis can be substantially increased by assuming linked loci. However, only 23% of the reviewed studies identified different potent risk-conferring haplotypes (9, 28-30) and explored the role of these haplotypes on cancer susceptibility. The sample size for each haplotype is too small for meta-analysis.

Despite these limitations, our results are statistically robust and yield important conclusions. There is a small association of the XPD Lys⁷⁵¹Gln polymorphism with risk for esophageal cancer and ALL. The Gln/Gln genotype is significantly associated with all of the cancers reviewed in this analysis. In addition, a slight but significant risk for bladder cancer is associated with the XPD Asp³¹²Asn polymorphism. No significant association was found for all of reviewed cancers with Asp³¹²Asn polymorphism. Although XPD polymorphisms cannot be considered as a crucial factor for cancer susceptibility, our results suggested that XPD is a highly suspected candidate gene without considering the role of environmental factors cautiously. In molecular epidemiology, due to the large number of candidate SNPs, there is a clear need to expedite genotyping by selecting and considering only a subset of all SNPs. Large, well-designed studies for haplotypes that, along with environmental factors, may influence susceptibility to cancer are recommended.